For decades, one principle governed how technology got built: linear instructions were cheap; intelligence was expensive. That split produced the entire software industry. Product companies encoded narrow problems into repeatable software and amortized the cost across thousands of customers. Services firms stitched those products together and added the customization layer products could not. The customer ended up with a stack of generic tools, held together by integration work and their own institutional knowledge.
The customization spectrum — off-the-shelf to bespoke — was just a measure of how much human integration had been added on top.
Then, AI happened.
The Way Things Were
A chain of generic products glued together by services firms worked. But every strength came with a hidden cost.
| Strength | Why it mattered | Weakness | Why it hurt |
|---|---|---|---|
| Best-in-class per slice | The best product for each part of your workflow | Workflow stitched across systems | Work moved across products. Context did not. |
| Designed once, used by many | Sophisticated software at a fractional cost | Built for the (hypothetical) "average" customer | The workflow fit no one exactly. Edge cases were a daily reality. |
| Documentable and trainable | The system was stable and visible. | Knowledge lived in PDFs and people | When the business changed, the workflow had to be rewritten, retrained, re-glued. |
| Predictable behavior | Each component did its narrow job consistently | Rigid by design | Anything outside the standard path required a custom project. The user bent to the system. |
| Clear interfaces | Users learned each product's screens and stayed productive | Integration lived in users' heads | A daily cognitive tax as users tracked down which system held which data or which screen produced which report. |
What AI Broke
Every "strength" was a workaround for intelligence being expensive. Here's how:
- Best-in-class slices: No single product could handle a full workflow intelligently.
- Design-once-use-many: Specifying per customer cost too much.
- Documentation: The system could not carry knowledge itself.
- Predictability: Reliability required uniformity.
- Clean interfaces: These were built to facilitate users because they had to compensate for systems that did not understand them.
As we wrote in January, AI increases the need for customization while dramatically reducing its cost. When intelligence gets cheap, the workarounds stop earning their keep.
What Works Now
A workflow that works in this environment should do four things.
Fix the actual problem. Don't deliver a "slice". Address the specific outcome the customer cares about. (Automating the wrong thing faster is not progress.)
Split work correctly. Volume, pattern matching, and execution are best left to the machine. Judgment, creative decisions, relationship contexts are still better left to humans.
Move fast. Humans work in two modes to direct the machine: allocation (what to do next) and imagination (how to push further). The machine does the work. This lets the work move faster, with human judgment guiding it.
Keep moving. The workflow is a continuous improvement engine. The firms bolting AI onto existing processes are not evolving workflows — they are preserving old ones with a new label on top.
Service as Software
The product/services divide is collapsing. We believe this is where the services mindset wins. Workflows have to evolve. A product mindset ships and exits. A services mindset stays embedded and keeps improving. Product economics with services posture is the new operating model. The firms that figure out how to price and staff for that combination will thrive.
Customization is Dead. Long Live Customization.
The old customization equation was a measure of how much human integration had been added to generic products. That constraint is gone. Reasoning models can now translate the same intelligence layer into role-specific outputs for different clients without touching the underlying system. AI didn't kill off customization. It changed where it happens and what it costs.
Firms still sticking to generic products or building at the data layer per client are running a pre-AI services model, just with higher software costs. Firms focused on customization at scale™ are running a different business entirely. If you're a tech services founder, it might be time to recalibrate where your firm stands so you don't run the risk of answering a question nobody is asking anymore.
We're taking the customization conversation further at our upcoming roundtable. If you are a tech services founder thinking through what your business looks like on the other side of this shift, RSVP here.